Authors: Chandravva Hebbi, Pradyumna Y M, Mamatha H R Abstract:This work focuses on building the database for handwritten Kannada words and make it open source as there are no publicly available datasets online. The dataset can be used as a benchmark dataset to evaluate the perfor-mance of various recognition, feature extraction methods, and the results of var-ious researchers. The dataset is rich in variations in writing styles because the data is collected from 370 different people. The words are extracted by the pro-jection profiles. Experimentation is done to recognize the words using Convolu-tion Neural Networks (CNN) with an accuracy of 93.20% and it is followed by category recognition with an accuracy of 91.90. An exhaustive dataset of 65378 words and 11 categories is built with experimentation